ElevenLabs Is Worth $22B — and It Answers Your Calls Now
ElevenLabs Is Worth $22B — and It Answers Your Calls Now
Most people who talk to ElevenLabs never know it. They are on hold with a bank, or sorting out a refund, or booking a dentist appointment, and the voice on the other end sounds like a person — patient, articulate, faintly warm. It is not a person. It is ElevenLabs, and it is now the most valuable voice AI company on the planet.
The company's backers reportedly value it at around $22 billion, TechCrunch reported on September 24, after a conversation with co-founder and CEO Mati Staniszewski at the Nrth entrepreneurship conference in Toronto. That figure, which traces to employee tender-offer talks Bloomberg first reported in July, roughly doubles the $11 billion valuation the company reached in its February Series D. It is a staggering price for a company that is just four years old.
The revenue, at least, is real. Staniszewski told TechCrunch the company is pacing at $600 million in annual recurring revenue, up from the $500 million it reported in May. More than 55 percent of that revenue now comes from what he called "classic enterprise" customers, with small and medium businesses, developers, and creators making up the rest. Four years after launching as a text-to-speech tool for narrators and YouTubers, ElevenLabs has become an infrastructure company — the voice layer underneath other people's customer service operations.
The scale of that layer is worth sitting with. Klarna runs first-line phone support for 35 million American customers on ElevenLabs' voices. Deutsche Telekom, Cisco, and Adobe use it. So do governments, in the United States and Europe. The company sells to creators too — audiobooks, dubbing, music — but the center of gravity has shifted. The biggest money in voice AI is no longer making a novelist's audiobook sound nicer. It is answering the phone.
The Turing test nobody passed — yet
Asked at Nrth how his earlier prediction was holding up — he had said at last year's TechCrunch Disrupt that audio models would be commoditized within a couple of years — Staniszewski was candid. The quality gap between models is still wide, he said. That will narrow within three to five years. His stated ambition for the interim is to be the first to pass "the Turing test for conversational AI."
That means more than fluency. "You need to combine intelligence, but you also need emotional intelligence," he told TechCrunch. "You need to understand the emotions of the other side, to be able to slow down or speak up. That hasn't yet been done." It is a notably different standard than the one the industry usually markets, which is about sounding human. Staniszewski's bar is behaving human — reading a caller's frustration and adjusting in real time.
Whether the Turing test metaphor is the right frame is debatable, and serious researchers would argue the original test was already passed by systems no one would call genuinely intelligent. But as a product target, it is revealing. ElevenLabs is not trying to win the race to generate a convincing podcast voice. It is trying to win the call.
The strangest rivalry in voice AI
There is an awkward tension running through the business. ElevenLabs' customers include companies that are turning into competitors. Decagon, a conversational AI platform, trained its voice product on ElevenLabs and now runs queries through its own models — effectively using the company's own customers as a funnel into a rival stack.
Staniszewski did not dodge it. "The lines become more blurry," he said, noting that the old categories — model company, platform company, application company — no longer hold. "In Anthropic's case, what was a model company is definitely a platform and increasingly a wide set of applications. I think this will continue." It is a graceful way of admitting that your biggest enterprise win today can be your most dangerous competitor tomorrow. The same pattern has played out across the AI industry: OpenAI's platform ambitions ate its partners, cloud providers moved up the stack, and every model company learned that the customer is always a potential rival.
Investors, at least, do not seem worried. A $22 billion valuation for a company pacing at $600 million in ARR implies roughly 37 times revenue — a multiple that says backers believe the voice layer will be bigger than the current business, not just a better version of it.
"I think there should be disclosure at this time"
The most interesting part of the interview had nothing to do with numbers. Asked whether businesses should tell callers they are talking to an AI, Staniszewski answered without hesitation.
"I think there should be disclosure at this time," he said. "Currently, people aren't used to it, and the common pattern is you don't want to feel cheated on that call."
The candor matters because it comes from the company whose technology makes the deception possible. ElevenLabs' voices are good enough that many callers never realize they are not speaking to a person — TechCrunch noted that most people encounter the company's technology "often without realizing it." The CEO's answer amounts to an argument for a rule that would, at least at the margins, make his own product's best trick less valuable. He added that the norm will change: "In five years, when everybody has their own agent working on their behalf, you'll be calling in and expecting an agent. Then I think we'll shift as a society."
For now, he suggested a practical compromise: give callers a choice. Facing a 30-minute wait for a human, offer the agent. "In almost all cases, they choose the agent and then they're surprised by how good the experience is."
The disclosure question is about to become a live regulatory issue, not just an etiquette debate. States are already experimenting with rules on synthetic media and AI impersonation, and consumer-protection law has little patience for businesses that let callers believe they are talking to a person when they are not. ElevenLabs' CEO has staked out the industry's most defensible position: disclose now, while the norm is forming, rather than be dragged into it later.
Margins are negotiable. Growth is not.
Asked about the company's gross margins — a pointed question, since ElevenLabs pays for models and inference and charges premium API rates — Staniszewski declined to discuss them in detail. But he made the strategic answer clear: "If we can pass on any savings to the customer, we do that... we don't mind the margins going lower to actually benefit together as the value gets created in the next five years."
This is the classic Amazon playbook applied to voice AI: accept thinner margins to make the product indispensable, and let scale do the rest. It is also a warning to competitors. ElevenLabs is willing to compress margins to win market share, which means anyone trying to undercut it on price — open-weight model providers like Mistral, or customers building on open-source stacks — will find the floor keeps moving down.
On the more delicate question of what goes into the models, Staniszewski said the company's advantage is less about the volume of training data than the annotation of it. Thousands of contractors annotate not just what was said in customer-service calls, but when people spoke, how they said things, and what emotions they used. The company brought in voice coaches to help detect accents accurately. It is painstaking, expensive work — the kind of thing that is hard to see in a benchmark and easy to hear in a call.
Governments, data residency, and the China question
TechCrunch pressed Staniszewski on a thornier issue: some of the open-weight models ElevenLabs' customers use are Chinese, and the U.S. government is a customer. The answer was careful. Each deployment is configured for the case at hand — open-weight, closed-source, or the customer's own fine-tuned model — with data residency kept in-country. In Poland, where the government uses ElevenLabs-powered agents to call patients with appointment reminders (18 percent of patients never show up), the models are optimized on Polish healthcare knowledge and the data stays in Poland.
It is a sign of how the business is maturing. ElevenLabs is no longer just an API for developers. It is negotiating with sovereign governments about where their citizens' voices get processed. That is enterprise software at its most serious — and it explains the 55 percent enterprise mix more than any single customer win.
On safety, Staniszewski drew a firm line: ElevenLabs does not build agents that can create more agents, every customer goes through KYC checks, and the company does not train the text and reasoning models that sit at the center of the industry's AI-safety debate. "We don't train the text models and the intelligence side of models, which is the core key of the debate," he said — a careful way of saying the scariest questions about autonomous AI are someone else's problem.
An IPO in 2028? Maybe.
Asked about reports that the company is targeting a 2028 IPO, Staniszewski would not confirm. "We'd love to create a company that stands the test of time. We are preparing the foundation to be able to do it in the next years. But whether we do it will depend on the time and place."
Pressed — "years is very vague" — he just laughed. It is the non-answer every private CEO gives, but the underlying point is serious: at $22 billion in the private market, ElevenLabs no longer needs the public markets for capital. An IPO would be an option, not a lifeline.
Whether the valuation holds is another question. Multiples of 37 times ARR are forgiving of exactly one thing: continued, explosive growth. The interview suggests Staniszewski knows it. He is betting that the next era of customer service is a phone call where both sides are machines — yours and theirs — and that ElevenLabs owns the voice layer in between. If he is right, $22 billion will look like a bargain. If the commoditization he predicts in three to five years arrives faster, the voice layer becomes a utility, and utilities do not command software multiples.
The phone is ringing either way. The question is who is answering — and whether you will be told.
Related reading: the AI-agent land grab is accelerating across consumer tech — see Instinct Raises $1B at a $10B Valuation for Its Personal AI Agent.
Sources: TechCrunch (Sept. 24, 2026, Connie Loizos interview at Nrth, Toronto); Bloomberg (July 2026, via TechCrunch/Startup Fortune, on tender-offer talks); Startup Fortune; AI Brief; RuntimeWire. ARR, enterprise mix, valuation, customer names, and margin/IPO commentary are company-reported or attributed to TechCrunch's reporting and are presented as such.
Comments
Post a Comment